What Are the Consequences of Real Earnings Management?

Katherine Gunny*
Haas School of Business
University of California, Berkeley CA 94720
Email: gunny@haas.berkeley.edu

January 2005

Abstract
This paper examines the consequences of four types of real earnings management. Using
financial statement data, I identify firms that engage in any of the following real earnings
management activities: (1) myopically investing in R&D to increase income, (2) myopically
investing in SG&A to increase income, (3) timing of income recognition from the disposal of
long-lived assets and investments, and (4) cutting prices to boost sales in the current period and
/or overproducing to decrease COGS expense. Then, I explicitly examine (i) the extent to which
real earnings management affects subsequent operating performance (as measured by both
earnings and cash flows), and (ii) whether market participants (investors and analysts) expect the
subsequent decline in performance. The empirical results are consistent with all four types of
real earnings management activities having a significantly negative impact on future operating
performance. Additionally, it appears that investors recognize the future earnings implications
of myopic investment in SG&A and cutting prices and/or overproducing to increase current
period income. The results are inconsistent with investors recognizing the future earnings
implications of myopic investment in R&D and the strategic timing of asset sales. The results
are consistent with analysts recognizing the future earnings implications of all four types of real
earnings management.

*
I am grateful for the guidance and support from my advisor, Xiao-Jun Zhang, and to Sunil Dutta, Qintao Fan,
Maria Nondorf, Shai Levi, James Powell and Mark Seasholes for their insights and suggestions. I would also like to
thank Tracey Zhang, Shimon Kogan, Gavin Cassar, Zvi Singer and Tatiana Fedyk for their helpful comments.
Errors or omissions are my responsibility.
1. Introduction
Earnings management can be classified into three categories: fraudulent accounting,

accruals management and real earnings management. Fraudulent accounting involves

accounting choices that violate GAAP. Accruals management involves within-GAAP choices

that try to obscure or mask true economic performance (Dechow and Skinner, 2000). Real

earnings management (RM) occurs when managers undertake actions that deviate from the first

best practice to increase reported earnings.1 This paper examine the extent to which real

earnings management affects subsequent operating performance and whether investors and

analysts recognize the consequences of real earnings management.

Schipper (1989) was one of the first to include RM in the definition of earnings

management. She describes earnings management as a purposeful intervention in the external

financial reporting process, with the intention of obtaining some private gain[a] minor

extension of this definition would encompass real earnings management, accomplished by

timing investment or financing decision to alter reported earnings or some subset of it.

Fraudulent accounting and accruals management are not accomplished by changing the

underlying economic activities of the firm but through the choice of accounting methods used to

represent those underlying activities. In contrast, RM is accomplished by changing the firms

underlying operations. Examples of RM include cutting prices towards the end of the year in an

effort to accelerate sales from the next fiscal year into the current year, delaying desirable

investment, and selling fixed assets to affect gains and losses, all in an effort to boost current

period earnings.

1
Conventional wisdom in prior studies is that managers prefer higher earnings, therefore the a higher stock price and
stock price is increasing in earnings (Fischer and Verrecchia, 2000). The focus of this study is on income
increasing RM, however there are situations in which the manager may benefit by deflating earnings. For example,
firms prior to a management buyout, vulnerable to an anti-trust investigation or seeking import relief may have
incentives to lower reported earnings (Fischer and Verrecchia, 2000; Perry and Williams, 1994; Watts and
Zimmerman, 1978; Jones, 1991).

1
 Even though accruals management may be less costly, with respect to future firm value,

there are several reasons managers may still engage in RM. First, aggressive accounting choices

with respect to accruals are at higher risk for SEC scrutiny and class action litigation, ex post.

Second, the firm may have limited accounting flexibility (i.e. limited ability to report

discretionary accruals). For example, accruals management is limited by the business operations

and by accrual manipulation in prior years (Barton and Simko, 2002). In addition, accruals

management must take place at the end of the year and managers face uncertainty as to which

accounting treatments the auditor will allow at that time. Operating decisions are controlled by

the manager, whereas accounting treatments must meet the requirements of auditors.

Prior studies provide strong evidence on the existence of RM. The use of RM by

managers is supported by Graham et al. (2004) who survey 401 financial executives about key

factors that drive decisions about reported earnings and voluntary disclosure. They report that

78% of the executives interviewed indicated a willingness to sacrifice economic value to manage

financial reporting perceptions. Furthermore, the extant empirical accounting literature confirms

the existence of RM to achieve various income objectives (see Section 2).

Given the extensive evidence on the existence of RM, this study examines the extent to

which RM affects subsequent operating performance. By definition, RM negatively impacts

future firm performance because the manager is willing to sacrifice future cash flows for current

period income. However, the extent to which various RM activities impact future operating

performance has not been addressed in prior literature. Using a matched sample of firms that did

not engage in RM, I find that four types of RM are associated with significantly lower future

earnings and cash flows after controlling for size, performance, level of accruals and industry.

Graham et al. (2004) document CFOs admitting a willingness to engage in RM as long as the

real sacrifices are not too large. However, my results suggest that RM activities are associated

with an economically significant decline in subsequent operating performance.

2
 Since the empirical results that all four types of real earnings management activities

negatively affects future operating performance, my second set of tests examines whether

investors and analysts recognize the consequences of RM. The results indicate that investors

recognize the future earnings implications of myopic investment in SG&A and cutting prices

and/or overproducing to increase current period income. However, the results are inconsistent

with investors recognizing the future earnings implications of myopic investment in R&D and

the strategic timing of asset sales. Analysts expectations, as reflected in forecasts of earnings,

appear to recognize the future earnings implications of all four types of RM.

Understanding the implications of RM is important not only to stakeholders of the firm

but also to accounting regulators. RM is one potential consequence of regulations intended to

restrict the discretion of accounting earnings management. For example, Ewert and Wagenhofer

(2004) develop an analytical model and demonstrate RM increases when tightening accounting

standards make accruals management more difficult. Although the present study does not

specifically address the tradeoff between accruals management and RM, examining the

consequences of RM provides general information relevant to assessing the costs and benefits of

accounting standards that may interact with the use of RM.

This paper makes the following contributions. First, it contributes to the literature on

earnings management. By undertaking a comprehensive examination of four types of RM, this

paper complements extant research investigating the consequences of earnings management.

Although there are several studies documenting whether RM occurs in various situations, the

existing literature provides little evidence of the affect of RM on firms subsequent operating

performance. This study provides a direct assessment of the impact of RM on future earnings

and cash flows. The purpose of this paper is not to identify specific motives for RM, but rather

to examine the consequences of RM. As a result, this study does not focus specifically on one

motive to engage in RM. I identify a broad sample of firms likely to have engaged in RM to

enhance the generalizability of the results. The empirical results demonstrate that subsequent

3
operating performance is negatively related to four types of RM.: (1) cuts in discretionary R&D,

(2) cuts in discretionary SG&A, (3) selling fixed assets, and (4) overproduction reflecting an

intention to cut prices or extend more lenient credit terms to boost sales and/or overproduce to

decrease COGS expense.

Second, this paper contributes to the literature on whether market participants recognize

the future earnings and cash flow implications of earnings management. Several studies

examine whether or not market participants identify and react to earnings management. These

studies examine either fraudulent accounting or accruals management (Dechow et al. 1996;

Sloan, 1996; Teoh et al. 1998; Bradshaw et al. 2001). In contrast, I examine whether investors

and analysts recognize the implications of RM. This study addresses the deficiencies in the

existing literature by examining the implications of RM.

Third, this paper contributes to the literature on earnings quality. Persistence of earnings

is an important part of the quality of earnings.2 In studies on financial statement analysis,

researchers are interested in how current or past earnings or earnings components aid in

forecasting future earnings or cash flows, both of which are central inputs in valuation models.

Examining the implication of RM on operating performance is important, given the significance

of future performance to the firm and its stakeholder. This paper shows that using empirical

measures to identify firms that engage in RM is incrementally informative about future earnings,

even after controlling for size, past performance, accruals and industry.

The remainder of the paper is organized as follows: Section 2 discusses the various types

of real earnings management and presents existing evidence. Section 3 develops testable

hypotheses. Section 4 describes the estimation models and the procedure to identify RM.

Section 5 describes the sample. Section 6 presents the results. Section 7 provides concluding

remarks.

2
Penman and Zhang (2002) define high quality earnings as sustainable earnings. Similarly, Francis et al. (2003)
state higher earnings quality signals that the earnings pattern is intrinsic and therefore sustainable, as opposed to
temporary and not attributable to fundamental firm characteristics.

4
2. Types of real earnings management (RM) activities and prior evidence

This study focuses on the following four types of real earnings management activities:
(1) decreasing discretionary R&D expense
(2) decreasing discretionary SG&A expense
(3) timing the sale of fixed assets to report gains
(4) overproduction reflecting an intention to cut prices or extend more lenient credit
terms to boost sales and/or overproduce to decrease COGS expense.3
There are other types of RM that are not examined in this study. For example, in

addition to the four RM activities listed above, Graham et al. (2004) document CFOs admitting

to delaying or cutting the travel budget and maintenance expense, postponing or eliminating

capital investments (to avoid depreciation charges), asset securitizations and managing the

funding of pension plans. This study does not propose an exhaustive list of all potential RM

activities. I focus on four commonly cited RM activities demonstrated to exist empirically in

prior research. A more comprehensive examination of the other types of RM is left to future

research.

2.1 Evidence on RM

2.1.1 Reduction of discretionary expense (R&D and SG&A)

Under current accounting rules, R&D expenditures must be charged to expense as

incurred because of the uncertainty of future benefits associated with investment in R&D (SFAS

No. 2, October 1974).4 As a result, a manager interested in boosting current period income could

choose to cut investment in R&D, particularly if the realization of the benefit associated with the

forfeited R&D project would benefit the firm in a future period, without hindering current period

earnings.

SG&A is included in the analysis because portions of this expense may be subject to

managerial discretion. GAAP does not consistently recognize intangible assets such as brands,

3
Managers can attempt to decrease COGS expense in any period by overproducing to spread fixed overhead costs
over a larger number of units as long as the reduction in per unit cost is not offset by inventory holding costs or any
increase in marginal cost in the current period.
4
FASB has permitted R&D to be capitalized only for certain kinds of software (SFAS 86).

5
technology, customer loyalty, human capital and commitment of employees as accounting assets,

all of which are created by expenditures on either SG&A or advertising.5 If the manager decided

to cut employee training programs intended to increase human capital and commitment of

employees, the economic consequence may not materialize in the short run but would in the long

run.

Several studies provide evidence that managers cut discretionary spending to achieve

earnings targets. Baber et al. (1991) provide evidence that R&D spending is significantly less

when spending jeopardizes the ability to report positive or increasing income in the current

period. Dechow and Sloan (1991) show that CEOs spend relatively less on R&D in their final

years in office. Bushee (1998) provides evidence consistent with institutional investors

mitigating the myopic investment problem. Bens et al. (2002) show that managers cut R&D and

capital expenditure when faced with earnings per share dilution due to stock option exercises.

Holthausen et al. (1995) find that managers do not cut R&D, advertising or capital expenditure

to increase managerial bonuses. Cheng (2003) provides evidence consistent with compensation

committees mitigating opportunistic reductions in R&D spending. With the exception of

managerial bonus incentives, the evidence is consistent with managers myopically investing to

achieve various income objectives.

2.1.2 Timing the sale of fixed assets to report gains

The timing of asset sales is a managers choice, and since a gain is reported on the

income statement at the time of the sale (the difference between the net book value and the

current market value), the timing of asset sales could be used as a way to manage reported

earnings. Bartov (1993) provides evidence consistent with managers selling fixed assets in order

to avoid negative earnings growth and debt covenant violations. Herrmann, Inoue and Thomas

5
Firms are not required to disclose advertising expense and if it is not separately disclosed it is included in SG&A,
therefore I do not require firms to disclose advertising expense to be included in the SG&A sample. Throughout the
paper any reference to SG&A expense refers to the aggregation of SG&A expense plus any advertising expense.

6
(2003) investigate Japanese managers use of income from the sale of assets to manage earnings.

They find that firms increase (decrease) earnings through the sale of fixed assets and marketable

securities when current operating income falls below (above) managements forecast of

operating income.

2.1.3 Overproduction

Roychowdhury (2003) points out that abnormally high production costs, for a given sales

level, is indicative of both: (1) sales manipulation due to abnormal price discounts, and (2)

COGS expense manipulation by overproduction. Sales manipulation refers to the behavior of

managers that try to increase sales during the current year in an effort to increase reported

earnings. By cutting prices (or extending more lenient credit terms) towards the end of the year

in an effort to accelerate sales from the next fiscal year into the current year, the firm is willing

to sacrifice future profits to book additional sales this period. The potential costs of sales

manipulation include loss in future profitability once the firm re-establishes old prices.

Managers can manipulate COGS expense in any period by overproducing to spread fixed

overhead costs over a larger number of units as long as the reduction in per-unit cost is not offset

by inventory holding costs or any increase in marginal cost in the current period. I use abnormal

production costs as one proxy for sales manipulation and COGS manipulation. Distinguishing

between these two types of RM by analyzing COGS expense and accounts receivables is

difficult because these items are susceptible to accruals manipulation.6 It is difficult to parse out

which effect is due to accruals manipulation and which is due to RM, therefore, I use abnormally

high production costs as a proxy for RM of sales or COGS.

Thomas and Zhang (2002) provide evidence consistent with managers overproducing to

decrease reported COGS, however, they can not rule out the possibility that the result is due to

6
For example, as pointed out by Roychowdhury (2003), if a manager decided to postpone a write-off of obsolete
inventory to decrease reported COGS, this decision would not affect production costs because the change in
inventories would be correspondingly higher. Production Costs = Cost of Goods Sold Expense + Change
Inventory.

7
adverse economic conditions. Roychowdhury (2003) develops empirical measures for RM of

discretionary expense and overproduction and finds that that managers trying to avoid reporting

losses, undertake RM. Firms suspected of RM exhibit unusually low cash flow from operations,

low discretionary expense and high production costs. The findings are consistent with managers

offering price discounts to boost sales, myopically investing and overproducing to decrease

COGS expense.

3. Hypothesis development

If a manager deviates from the optimal level of activity and engages in RM, then

presumably there would be long run economic consequences. RM negatively impacts future

firm performance because the manager is willing to sacrifice future cash flows for current period

income. Since the future firm performance in the absence of RM is not observable, I identify a

control sample of non-RM firms matched on industry, performance and accruals decile. I

would expect RM firm-years to have lower subsequent operating performance compared to the

control sample. Examining the subsequent operating performance between the RM firms and the

control firms will allow me to assess the extent to which various RM activities impact future

operating performance.

Alternatively, if the RM firms operating performance in subsequent years is

indistinguishable from the control firms, then this result would be consistent with two

explanations: (1) the identified RM firms did not engage in real activities manipulation (2) the

signaling hypothesis is true, that is, RM reveals managers private information about future firm

performance. Survey data by Graham et al. (2004) indicate that CFOs, knowing RM may hinder

future performance, engage in RM in the hope that future earnings growth will offset current

RM. Several papers find evidence consistent with earnings management being positively

associated with the managers expectation of future performance (Subramanyam, 1996; DeFond

and Park,1997; Altamuro et al., 2003).

8
Hypothesis 1: Compared to a performance, industry and accruals-matched sample, RM
firm-years have relatively lower subsequent operating performance (both
return on assets and cash flow from assets).

Next, I examine the behavior of investors, in an effort to assess whether investors

understand the earnings implications of RM. The evidence as to whether investors recognize

earnings management (specifically accruals) is mixed. On the one hand, investors seem to not

fully see through earnings management as reflected in abnormal accruals (Sloan, 1996; Xie,

2001). On the other hand, in the banking and insurance industries, loan and policy loss reserves

are two major accounts subject to management discretion and investors do understand the

information in these accruals (Wahlen, 1994; Beaver and Engel, 1996; Liu et al., 1997; Beaver

and McNichols, 2001). To ascertain whether investors detect and, therefore, react to RM, I

examine the extent to investors incorporate the future earnings implications of RM into stock

prices.

Hypothesis 2: After controlling for known risk factors, there is no association between
firms subsequent year returns and current-year RM.

To the extent investors detect RM, I would expect stock market prices to efficiently

impound the information about RM for future earnings. Similarly, to the extent analysts detect

and understand the implications of RM, I would expect no association between RM and

subsequent forecast errors.7

Hypothesis 3: There is no association between subsequent analysts forecast errors and

current year RM.

7
Financial statement analysis texts oftentimes implore analysts to examine various transactions related to RM. For
example, Palepu, Healy and Bernard (2004) suggest that the analyst should ask has the firm structured its activities
(such as R&D, design, manufacturing, marketing and distribution, and support activities) in a way that is consistent
with its competitive strategy? Additionally, they identify selling assets to realize gains in periods when operating
performance is poor as a potential red flag that should lead the analysts to examine this item more closely.

9
4. Identification of RM and estimation models

First, prior literature is used to develop models to calculate the expected (i.e. normal)

level of four accounts representing operational activities linked to RM: R&D expense, SG&A

expense, gain (loss) on asset sales, and production costs. The abnormal level of each measure

for every firm-year is calculated as the actual value minus the estimated normal level. Second,

in an effort to increase the power of correctly identifying firms that engage in RM, I restrict the

sample to firms with low accruals flexibility. Theory and empirical papers have demonstrated

that firms with limited flexibility to engage in accruals management are more likely to engage in

RM. The intersection of firms with abnormal account levels consistent with RM and low

flexibility to engage in accruals management are identified as firms suspected of engaging in

RM. For example, a firm with limited accounting flexibility and in the lowest abnormal R&D

expense quintile is identified as having engaged in myopic investment in R&D. A firm with

limited accounting flexibility and in the highest abnormal asset gain quintile is identified as

having engaged in strategically selling assets to realize gains.

4.1 Proxy for limited ability to engage accruals management

An analytical paper by Ewert and Wagenhofer (2004) show that in the face of tightening

accounting standards, managers substitute into using RM which is costly and reduces firm value.

Using a simultaneous equation approach, Zang (2003) predicts and finds that firms with higher

levels of previous earnings management are more likely to use RM relative to accruals

management. Barton and Simko (2002) provide evidence that managers ability to optimistically

bias earnings with accruals management decreases with the extent to which net assets are already

overstated on the balance sheet.

Consistent with Barton and Simko (2002), I use the beginning balance of net operating

assets (i.e. shareholders' equity less cash and marketable securities, plus total debt) to identify

firms with limited accounting flexibility. The rationale is that the articulation between the

income statement and the balance sheet ensures that biased assumptions reflected in earnings are

10
also reflected in net asset values. As managers attempt to improve current period earnings by

deferring costs into the future, their ability to manage future earnings upward is restricted due to

reversals of previously deferred costs.8 Therefore, the balance sheet accumulates the effects of

previous accounting discretion.

4.2 Estimation models

4.2.1 The normal level of R&D expense

The normal level of R&D expense is estimated using the following model:

RD t RD t 1
= 0 + 1 + 2 INT t + 3 Q t + 4 CX t + 5 MV t +
A t 1 A t 1

(1)

Where: RD = Research and Development expense deflated by lagged total assets [#46/#6]
A = Total assets [#6]
INT = Internal Funds [(#18+#46+#14)/ #6]
Q = Tobins Q: firms market value divided by the replacement cost of its assets [((#199*#25)+#130+#9+#34)/#6]

CX = Capital Expenditures [#30/#6]

MV = Log of Market Value of Equity [log(#199*#25)]

Equation (1) is based on Berger (1993) who develops an expectations model for the level

of R&D intensity. The model is estimated for every year (1988-2000) and industry (48

industries based on the classification system developed by Fama and French, 1997). The

independent variables are designed to control for factors that influence the level of R&D

spending. The prior years R&D serves as a proxy for the firms R&D opportunity set and the

coefficient would be expected to be positive. Internal funds is a proxy for reduced funds

available for investment. Tobins Q is a proxy for the marginal benefit to marginal cost of

installing an additional unit of a new investment. Capital expenditure is a proxy for the

competition for resources between capital expenditure and R&D. The adjusted R-squared for the

R&D estimation model is generally above 90% for all industry-year combinations.

8
Barton and Simko (2002) find that higher levels of beginning-of-the-period net NOA scaled by sales are negatively
associated with the probability of at least meeting the consensus analyst forecast for the current period. Hansen
(2004) finds similar results using last periods earnings as the benchmark. Kasznik (1999), using change in prior
accruals as the proxy for flexibility, shows that managers who overestimate earnings manage reported earnings
toward their forecasts to a greater extent when they have more accounting flexibility.

11
4.2.2 The normal level of SG&A expense
The normal level of SG&A is estimated using the following model:

SGAt S S S S
log( ) = 0 + 1 log( t ) + 2 log( t ) * DDt + 3 log( t 1 ) + 4 log( t 1 ) * DDt 1 +
SGAt 1 S t 1 S t 1 S t 2 S t 2

(2)

Where: SGA = SG&A + Advertising expense

S = Sales [#12]
DD = Indicator variable equal to 1 when sales revenue decreases between t-1 and t, zero otherwise

The estimation of SG&A, equation (2), incorporates controls for the sticky cost

behavior shown by Anderson, Banker and Janakiraman (2003). The model is estimated by

industry and year. In particular, costs are sticky if the magnitude of a cost increase associated

with increased sales is greater than the magnitude of a cost decrease associated with an equal

decrease in sales. The general theory is that managers tradeoff the expected costs of maintaining

unutilized resources during periods of weak demand with the expected adjustment costs of

replacing these resources if demand is restored. Not including this element into the SG&A

expectations model may lead to underestimating (overestimating) the response of costs to

increases (decreases) in sales.9 Similar to Anderson et al. (2003) the adjusted R-squared is

generally above 50% for all industries and years.

4.2.3 The normal level of gains on asset sales

The normal level of gain on asset sales is estimated using the following model:
GainA t = 0 + 1 ASales t + 2 ISales t + 3 log( S ) t + 4 Growth t + (3)

Where: GainA = Income from asset sales deflated by beginning of the year stock price [Data213/Data199]
ASales = Long-lived assets sales / market value at the beginning of the year [data107/(#199*#25)]
ISales = Long-lived investment sales / market value at the beginning of the year [data109/(#199*#25)]
S = Sales [#
Growth = The percentage change in sales for the current period [Salest Salest-1]/Salest ]

9
The ratio form and log specification improves the comparability of the variables across firms and alleviates
potential heteroskedasticity due to large differences in the size of firms.

12
 Equation (3) is based on Bartov (1993) and augmented by variables in Herrmann et al.

(2003) shown to influence the level of gain on asset sales. Introducing asset sales as an

explanatory variable in equation (3) requires that the relation between income from asset sales

(GainA) and asset sales (ASales) and investment sales (ISales) be monotonic. Therefore, the

variables are transformed to make the relationship monotonic, so when income from asset sales

is negative, asset sales and investment sales enter the regression with negative signs, therefore a

positive coefficient would be expected.10 Total sales (S) is included in the regression to control

for any size effects. Growth is included in the regression to control for the expectation that

growth firms are less likely to recognize gains because they are in a period of expansion. The

average adjusted R-squared is around 30%.

4.2.4 The normal level of production costs

The normal level of production cost is estimated using the following model:
PRODt 1 S S t S t 1 (4)
= 0 + 1 t + 2 + 3 +
At 1 At 1 At 1 At 1 At 1

Where: PROD = COGS + INV

COGSt = Cost of goods sold expense [#41]
INVt = Inventory increase/decrease [#303]
St = Sales [#12]
At = Total assets [#6]

This model is presented in Dechow et al. (1998) and implemented by Roychowdhury

(2003) to estimate the normal level of production. The model is estimated for every year and

industry. Abnormally high production costs may indicate overproduction to decrease COGS or

sales manipulation. The adjusted R-squared is generally above 90% for all firm-year

combinations.

4.2.5 Alternative estimation models as a robustness check

I implement supplemental analysis using alternative expectation models for R&D

expense, SG&A expense and gain (loss) on asset sales. First, I model R&D and SG&A expense

10
Income from asset sales is increasing (decreasing) in asset sales when income is positive (negative).

13
(deflated by assets) just as a function of sales as described by Dechow et al. (1998) and

implemented by Roychowdhury (2003). These alternative specifications yield similar results,

however I choose to use the expectations models that control for more factors beyond current

period sales to alleviate concerns about the strategic considerations that influence the level of

discretionary spending. Second, I estimate the normal level of income from asset sales as

income from asset sales minus the median for the corresponding industry and year (based on

Herrmann et al. 2003). The results are similar using this specification.

5. Sample selection and data

5.1 Data and sample selection

The sample consists of all firms with available financial data from COMPUSTAT

industrial, full-coverage and research files and stock returns and size portfolio returns from

CRSP. Firms in the financial industry (SIC 6000-7000) and utility industry (SIC 4400-5000) are

excluded because they operate in highly regulated industries with accounting rules that differ

from those in other industries. Only firms that are traded on the New York Stock Exchange,

American Stock Exchange, and NASDAQ are included in the sample. The sample includes

annual data for firms covering years from 1988 to 2000.

The sample is restricted to pre-2000 data so there are three years of subsequent earnings

to examine. The sample is restricted to post-1987 data, because data on income from asset sales

is not available on COMPUSTAT prior to 1987, and to facilitate the consistent calculation of

accruals via the statement of cash flows.11 The final sample consists of 32,402 firm-year

observations with required financial statement variables and returns data. The four RM samples

are a subset of the full sample with data available to calculate the normal level of each RM

11
Hribar and Collins (2002) show that the accruals calculated from cash flow numbers reported by firms under
SFAS 95 are less susceptible to the contaminating influences of acquisitions, mergers, and divestitures than accruals
calculated from the balance sheet. SFAS 95, the standard governing the preparation of the statement of cash flows
took effect in fiscal 1988.

14
activity. The R&D, SG&A, Asset and Production RM sample consists of 19,366, 24,628, 30,422

and 30,125 firm-year observations, respectively.

5.2 Identification of RM

To identify years when firms were likely to have engaged in RM, I first calculate the

abnormal level of the four types of activities associated with RM for each firm-year. Next, I

calculate the level of previous earnings management (NOAt-1) for each firm-year. NOAt equals

net operating assets (shareholders equity less cash and marketable securities plus total debt)

divided by lagged assets. DeFond (2002) points out in his review of Barton and Simko (2002)

that there are likely to be systematic differences in the ratios across industries, as well as firm-

specific effects on the ratios, that are unrelated to whether net assets are overstated. In an effort

to mitigate this bias, I rank firms based on NOAt-1 by year and industry.

Barton and Simko (2002) use quarterly data, therefore in an effort to show the same

discretionary accruals pattern holds annually, I replicate their analysis using annual data. Table

1 presents the mean and median prior cumulative abnormal accruals across quintiles of the

beginning balance of net operating assets. NOAt-1 is net operating assets (i.e. shareholders' equity

less cash and marketable securities, plus total debt) at the beginning of quarter t, scaled by sales

for quarter t-1. Abnormal accruals are estimated for firm i in year t using the residual from the

modified Jones model, estimated by two-digit SIC code. Table 1 reveals, similar to Barton and

Simko (2002) who use quarterly data, that prior cumulative abnormal accruals across quintiles of

NOAt-1 are larger for higher NOAt-1 quintiles. The results are consistent with the notion that the

balance sheet accumulates the effects of previous accruals management and the level of net

operating assets partly reflects the extent of previous earnings management.

Table 2 reports the frequency of firms by quintile of each type of abnormal RM activity

in year t and quintile of net operating assets in year t-1. The column on the right presents

descriptive statistics by quintile of each abnormal RM activity measure. The lower right cell in

each frequency panel indicates the number of firms suspected of RM. I identify firms suspected

15
of engaging in cutting R&D activities beyond an optimal level (R&D RM) as firms in the lowest

abnormal R&D quintile in year t and the highest abnormal NOA quintile in year t-1 (976 firm-

years). I identify firms suspected of engaging in cutting SG&A activities beyond an optimal

level (SG&A RM) as firms in the lowest abnormal SG&A quintile and the highest abnormal

NOA quintile in any given year 1,080 firm-years). I identify firms suspected of engaging in the

timing of asset sales (Asset RM) as firms in the highest abnormal gain on asset quintile and

highest NOA quintile in any given year (1,150 firm-years). I identify firms suspect of either

sales manipulation or COGS manipulation (Production RM) as firms in the highest abnormal

production quintile and highest NOA quintile in any given year (898 firm-years).

Panel A shows 976 firm-years where the firm is suspected of R&D RM. Panel B shows

that 1,080 firm-years are suspected of SG&A RM. Panel C reveals that 1,150 firms are

suspected of Asset RM. Panel D shows that 898 firm-years are suspected of Production RM.

Unreported results reveal that the mean level of accruals deflated by average assets is not

significantly higher for the RM firm-years than the rest of the sample. For example, accruals are

.045 for the 976 R&D RM firm-years and .046 for the rest of the R&D sample (not

statistically different). Scaled abnormal R&D (SG&A) expense for the lowest quintile is 13%

(24%) compared to 0% (0%) for the middle quintile. Scaled abnormal gain on asset sales

(production cost) is 45% (38%) for the highest quintile compared to 3% (4%) for the lowest

quintile.

5.1.2 The inherent difficulty of identifying earnings management

Given the inherent difficulty in identifying earnings management without knowing the

managers true intention, a criticism of the earnings management literature is that any earnings

management identified may be a result of an omitted variable or may be capturing behavior other

than intentional manipulation. The same is true for the present study, however, I try to mitigate

these concerns in a few ways. First, firms likely to have engaged in RM are identified in the

two-step process explained above: (1) abnormal accounting levels consistent with RM and (2)

16
less accounting flexibility. Limiting the sample to firms with less accounting flexibility should

increase the power of detecting RM. Additionally, in the examination of subsequent return on

assets, the regression controls for several other factors that may influence future profitability.

5.3 Descriptive statistics

Table 3 presents descriptive statistics for the full sample and the sample of RM firm-

years. The 3,129 RM firm-years contain 2,280, 730, 112 and 7 firm-years that engage in one,

two, three and four types of RM, respectively. The data reveals that RM firm-years have mean

total assets of 1,243 million, similar to the full sample with total assets of 1,331 million. Mean

total sales for the RM sample are smaller 772 million compared to 1,407 million for the full

sample. The mean market capitalization of the RM firms, at around 902 million, is smaller than

that of the full sample, 1,808 million. RM firm-years are less profitable (return on assets, cash

flow from assets) compared to the full sample.

Table 3 (cont.) reports the industry composition by RM sub-sample. I use the

classification scheme devised by Fama and French (1997). They categorize firms using groups

of four-digit SIC codes to ensure that similar firms are grouped together. The most represented

industry is business services at 13.1% of the full sample, and next is electronic equipment at

8.2% of the sample. Approximately 23% of the R&D RM sample comes from business services;

the next most represented industry is pharmaceutical products. The industry compositions of the

SG&A and Asset RM samples are similar to the full sample.

The most represented industry in the Production RM sample is pharmaceutical products.

This may seem peculiar given manufacturing industries should be primarily responsible for the

abnormal production costs due to COGS manipulation. However, overproduction is a proxy for

both COGS manipulation and sales manipulation, and it may be the case that non-manufacturing

firms are more aggressive at offering price discounts (sales manipulation) than manufacturing

firms. For example, in August 2004, Bristol-Myers, a large pharmaceutical manufacturer, agreed

17
to pay $150 million to settle SEC charges of accounting fraud, largely stemming from channel

stuffing, or enticing wholesalers to buy excess inventory.12

6. Results

6.1 Operating performance after real earnings management

The operating performance of firms that engage in RM is tested separately for each

sample: R&D RM, SG&A RM, Asset RM and Production RM (Hypothesis 1). Two methods are

employed. First, in an effort to control for both performance and accruals manipulation, I adopt

a matched sample technique. For every RM firm, a control firm matched on operating

performance, industry and accruals decile in year t is identified. Next, I compare performance in

the subsequent three years. Finally, future ROA is modeled as a function of current ROA, an

indicator variable for RM, an interaction variable between RM and ROA and several control

variables.

6.1.1 Operating performance - univariate analysis

In this section changes in operating performance are examined in the subsequent three

years between the RM firms and a matched control firms. In this setting, I attempt to control for

three main issues that may influence future firm performance: the level of accounting accruals,

the mean reversion in accounting data, and industry membership.

Two measures are used to capture operating performance: return on assets and cash flow

from assets. Return on assets (ROA) is defined as earnings before extraordinary items divided

by average total assets. This measure is commonly used in the accounting literature. The costs

of RM include the possibility that cash flows in future periods are affected negatively by the

actions taken during this period to increase earnings. Therefore, the second accounting

performance measure is cash flow from operations on assets (CFO).

12
The Wall Street Journal, August 5, 2004, Bristol-Myers Settles SEC Fraud Case by Barbara Martinez.

18
 Extensive accounting research documents that firms with high accruals are more likely to

experience future earnings problems. Therefore, to ensure the results are robust to the accrual

anomaly, I match on accruals decile in year t. Next, to control for the underlying economic

factors as well as the mean reversion in accounting data, Barber and Lyon (1997) suggest that

past performance adjusts for these effects. Many studies match on size, but doing so assumes

that operating performance varies by size (Fama and French, 1997) and Barber and Lyon (1997)

suggest that size is not essential for detecting abnormal operating performance. Kothari et al.

(2002) contend that performance matching is useful when variables of interest are correlated

with performance. They suggest that performance matching is critical to designing well-

specified tests of earnings management. Finally, the RM firms are matched on industry

membership to control for any industry wide shocks and conservative accounting inherent in the

industry.

Each RM firm is matched with a non-RM firm by year, industry and performance. Since

the analysis examines four different types of RM, I construct a control sample for each separate

RM sample. I require that the operating performance (ROA or CFO) in year t be within 90

110% of the operating performance of the RM firm and in the same accruals decile and same

four digit SIC. If no firm is within this confidence interval, the process is repeated but increase

the sample of potential matches by identify all firms in the same accruals decile in a similar three

digit SIC. If this does not produce a match I identify all firms with a similar two digit SIC then

if necessary by one digit SIC. If a match still cannot be found, I match performance within the

designated threshold without regard to SIC. In all cases, the control firm is in the same accruals

decile, and within 90110% of the RM firm.13 The majority of matches are made at the 2 digit

and 3 digit SIC level.

13
In some cases, the matching procedure does not identify a control firm, because there is no other firm within the
90% to 110% performance threshold and in the same accruals decile. Therefore, I remove these RM firms from the
analysis. As a robustness check, instead of removing these RM firms from the analysis, I identify a control firm
closest in performance without regard to the 90% to 110% performance criteria. The results are qualitatively
similar.

19
 The abnormal performance in the subsequent three years is equal to the difference

between the performance of the RM firm and the contemporaneous performance of its match

control firm as follows: Abnormal ROAj,t = ROAi,t Matched ROAi,t (5)

Abnormal CFOj,t = CFOi,t Matched CFOi,t (6)
Table 4 presents the univariate analysis. Panel A reports abnormal operating

performance for the R&D RM sample. In year t, the matching year, abnormal operating

performance is not significantly different from zero which one would expect given the

performance matching criteria. Consistent with R&D RM having an economically significant

consequence, operating performance (both means and medians) is lower in each of the

subsequent three years. Mean (median) abnormal ROA is 1.15% (4.19%) lower in the

subsequent year for firms that engaged in myopic investment in R&D. The evidence is

consistent R&D RM having a significantly negative impact on both earnings and cash flows.

Panel B reports abnormal operating performance for the SG&A RM sample. Abnormal

ROA in the subsequent three years is significantly negative for the SG&A RM firm-years. The

biggest impact on both ROA and CFO occurs in t+1. Although CFO is lower for SG&A RM

firm-years, it is not significantly lower in any of the subsequent three years. The results indicate

that SG&A RM is negatively related to future earnings but no significant cash flow effect,

compared to the matched sample.

Panel C presents the results for the Asset RM sample, abnormal ROA and CFO is

negative in the subsequent three years. Median abnormal ROA is significantly negative in year

t+2 and t+3. Median CFO is significantly negative in year t+1 and t+2. The results suggest that

Asset RM has a negative impact on cash flows immediately, but return on assets is not affected

until later. One interpretation could be that the firm no longer has to report depreciation expense

associated with the strategically sold asset, therefore earnings are not significantly lower in year

t+1. However, the foregone future cash flows associated with Asset RM are apparent

immediately.

20
 Panel D reports abnormal operating performance for the Production RM sample.

Abnormal ROA in the subsequent three years is significantly negative, whereas abnormal CFO is

only significantly negative in year t+1. Mean (median) abnormal ROA is 3.26% (1.27%) lower

in the subsequent year for firms that engaged in production RM and mean (median) abnormal

CFO is 1.79% (1.46%) lower in the subsequent year. The evidence is consistent R&D RM

having a significantly negative impact on both earnings and cash flows.

Table 5 reports the results for abnormal changes in operating performance after RM. All

types of RM, except Asset RM, appear to be associated with lower changes in ROA in year t+1,

evening after controlling for the accrual anomaly and industry membership. However, in year

t+2 and t+3 both the changes in ROA and CFO of RM firm-years are not distinguishable from

the matched sample (except for abnormal change in CFO for Asset RM firms is significantly

positive in year t+3). Taken together, the findings reported in Table 4 and Table 5 support the

hypothesis that RM has an economically significant impact on subsequent operating

performance. Specifically, all four types of RM activities are associated with lower future

performance compared to non-RM firms matched on performance, accruals and industry.

5.1.2 Operating performance multivariate analysis

There may be other factors (besides industry, accruals decile and performance) that

influence future operating performance. In an effort to check the robustness of the results to

alternative explanations, I add additional control variables that may influence operating

performance.14 In this section, the regression controls for performance, size effects, growth

opportunities, the life cycle of the firm and performance. ROA in year t+2 is also examined,

because the future performance effects of RM are likely to take some time to impact future ROA.

ROAt+i = 0 + 1LOGASSETt + 2BTMt + + 3ROAt + 4RETURNt + 5PORTACCt

+ 6I_RMnt + 7I_RMnt*ROAt + t+1 (7)

14
This model is based on Bens et al. (2002) who examine the real cost of awarding employee stock options. The
model is augmented to control for accruals and RM.

21
Where: i = 1 to 2
n = 1 to 4
ROA = return on assets in year tin year t (#123/ beginning of the year total assets)
LOGASSET = the natural logarithm of total assets
BTM = the book value of equity divided by the market value of equity
RETURN = the one year holding period return on an investment in firm j's common stock
PORTACC = the portfolio ranking of accruals, converted to a [0,1] scale

In this section, model (7) is estimated on the reduced sample of firm-years in the high net

operating asset quintile in an effort to control for any bias potentially induced by identifying RM

firm-years based on high net operating assets. Running the regression on the reduced sample

ensures the results are robust to any potential bias due to high net operating assets15.

I include past ROA to control for the time series properties and performance of return on

assets. LOGASSET controls for any size effects. BTM controls for growth opportunities and/or

the life cycle of the firm. In the context of R&D, SG&A and Asset RM, controlling for the life

cycle is important, given the maturity hypothesis which predicts that as firms mature, they

experience a decline in their investment opportunity set. I also include RETURN to control for

the association between stock performance and future earnings. Kothari and Sloan (1992) show

that current stock prices predict future earnings positively. Finally, PORTACC controls for the

accrual anomaly documented by Sloan (1996).

The coefficient estimates for model (7) are presented in Table 6 and 7. The control

variables, current ROA and RETURN, are significant and with the predicted sign. With the

exception of the R&D RM model, BTM is significantly negative. The coefficient estimate on

PORTACC is significantly negative in every model consistent with the evidence that higher

levels of accruals contribute negatively to future ROA.

Even after controlling for the time-series property of ROA, size effects, growth

opportunities, and performance, identifying R&D, SG&A and production RM are incrementally

informative at explaining future ROA. However, the interaction term, representing the

15
Qualitatively similar results are obtained when estimating model (7) using the full sample.

22
persistence of ROA, is significantly positive for R&D RM firm-years. The persistent of ROA

for the Asset and Production RM sample is significantly negative.

Since it may be the case that future performance effects of RM take longer than one

period to materialize, I estimate model (7), using ROAt+2 as the independent variable, to assess

the impact of ROA on two years ahead ROA. The persistence of ROA on two years ahead ROA

for the R&D RM sample is negative but not significant, however, the mean effect is significantly

negative. Although SG&A RM firms have worse future firm performance, represented by the

I_RM2 indicator variable, the persistence of ROA is significantly positive for ROAt+1. Overall

the results of the multivariate analysis suggests that identifying all four types of RM are

incrementally informative about future earnings, even after controlling for information in past

earnings and several control variables.

6.2 Does the stock market recognize real earnings management Mishkin Test

Following Mishkin (1983) and Sloan (1996), I test whether the stock market is efficient

in impounding the information contained in RM for future earnings. First, I estimate the relation

between RM and future earnings. Since the univariate results reveal that, at least partially, the

performance consequences of RM materialize in the subsequent year, the analysis focuses on

year t+1. Second, the relation between RM and future earnings implicit in security prices is

estimated. A comparison of these historical and market-inferred weights using the Mishkin test

indicates whether investors correctly identify RM and its importance for future earnings. The

earnings forecasting equation in Sloan (1996) is extended to incorporate the implications of RM

for future earnings as follows:

EARNt+1=0 + 1aCFOt + 1bACCt + 2I_RMnt + 2aCFOt*I_RMnt + 2bACCt*I_RMnt +t+1 (8)

SARt+1 = 0 + 1 (EARNt+1 0 1a* CFOt 1b* ACCt 2I_RMnt

2aCFOt*I_RMnt 2bACCt*I_RMnt) + t+1 (9)

23
 The RM group is compared to the rest of the sample, therefore RM takes the value of 1 if

the firm engaged in RM and the value of 0 otherwise.16 Equation (8) is the forecasting equation;

the coefficient 1a and 1b captures the persistence of cash flows and accruals, respectively,

while 2a (2b) captures the differential persistence factor for cash flow (accruals) between the

RM firm-years and the rest of the sample. Equation (9) assumes that the market reacts to

unexpected earnings conditioned on last years earnings and estimates the weights that the

market assigns to the earning components in forecasting future earnings. Comparing coefficients

across equations tests whether the market prices cash flows and accrued earnings efficiently in

either the non-RM or the RM group.

The equations are estimated jointly using an iterative generalized nonlinear least squared

estimation. The system is run twice. First, the system is run with no constraints. Second, to test

whether the weight on the earnings components is the same between the forecasting and pricing

equation the system is run again imposing the coefficient constraints being tested. The equality

of the coefficients across equation (8) and (9) is tested using a likelihood ratio statistic which is

distributed asymptotically chi-square (q):

2*n*Ln(SSRc/SSRu). q is the number of constraints imposed by market efficiency; n is

the number of observation in each equation; SSRc is the sum of squared residuals from the

constrained weighted system; and SSRu is the sum of squared residuals from the unconstrained

weighted system.

5.2.1 Results of the Mishkin test

Table 8 reports the results from the Mishkin test. Consistent with Sloan (1996), it

appears that investors underestimate the persistence of cash and overestimate the persistence of

accruals in each sample. For example, in Panel B, the coefficient on CFO (ACC) in the

16
I compute the results of the Mishkin test on full sample and the results are qualitatively similar.

24
forecasting equation is 0.80 (0.66), whereas the coefficient from the pricing equation is 0.69

(0.75).

Testing 2a= *2 a (2b= *2 b) indicates whether the market recognizes the differential in

the persistence of cash flows (accruals) between the RM group and the non-RM group.17 Panel

A of Table 8, reports the results of the Mishkin test for the R&D RM sample. The coefficient on

the R&D RM indicator variable is -0.026 indicating that R&D RM firms are associated with

lower future earnings, whereas the market perceives the weight to be 0.024. The likelihood ratio

statistic indicates that the difference between the forecasting and pricing equation is significantly

different (ratio statistic 7.29). The differential persistence factor for cash earnings is 0.010 (not

significant), while the market perceives it to be .118 (significant with one tail). The likelihood

ratio statistic indicates that the market does not appear to misestimate the persistence in cash of

R&D RM. However, the difference is significant using a one tail test. Similarly, the market

efficiently prices the accrual component of R&D RM firms. Taken together, it appears the

market overestimates the contribution of R&D RM firm-years to future earnings. However, the

market correctly prices the persistence of the cash and accruals components of these earnings.

Panel B of Table 8, reports the results of the Mishkin test for the SG&A RM sample.

The differential persistence factor for cash earnings is insignificant and zero, however, the

market perceives it to be 0.152 (with a significant likelihood ratio of 5.97). The differential

persistence factor for accrued earnings is 0.05, while the market perceives it to be 0.05, not

statistically different. It appears the market overestimates the persistence in cash flows

associated with SG&A RM firm-years, although the market does recognize the lower persistence

of accruals and the lower future earnings for these firm-years.

Panel C of Table 8, reports the results of the Mishkin test for the Asset RM sample. It

appears the differential persistence factor for both cash and earnings is equal across the

17
Since the interest is whether investors recognize the differential in the persistence of cash flows and accruals
between the two groups, I just report the likelihood ratio statistic for that test.

25
forecasting and pricing equation. The indicator variable for Asset RM firm-years is significantly

negative in the forecasting equation but significantly positive in the pricing equation. It appears

the market assigns a higher weight to Asset RM firms than is justified given the relationship

between Asset RM firms and future earnings (with a significant likelihood ratio statistic of 7.46).

Overall, it appears the market recognizes the persistence in accruals and cash flows associated

with Asset RM but not the mean effect.

Panel D of Table 8, reports the results of the Mishkin test for the Production RM sample.

The differential persistence factor for cash earnings is positive, 0.075, while the market perceives

it to be 0.170 (with an insignificant, two-tailed, likelihood ratio of 2.07). The differential

persistence factor for accrued earnings is .169, while the market perceives it to be 0.376 (with

a significant likelihood ratio of 3.62). Interestingly, the results indicate that the market

underestimates the persistence in accruals for the Production RM firms. Overall, the Mishkin

test indicates that the market overestimates the one-period ahead earnings of Production RM

firm-years and the persistence of cash flows (although only significant two-tailed). However,

the market underestimates the persistence in accruals.

5.3 Additional tests

5.3.1 Future returns and real earnings management

The preceding section used the Mishkin test to demonstrate whether the stock market

prices information about RM. The coefficients are estimated from a set of contemporaneous

observations (throughout the sample period), hence the procedure suffers from a foresight bias.

Because the models use future information the market did not have when setting prices, these

regressions do not provide a valid test of market efficiency.18 As a result, I will provide

additional tests in an effort to mitigate these potential biases. I estimate the following cross-

sectional OLS regression for each of the 13 years in the sample:

18
See Beaver and McNichols (2001); Kraft, Leone and Wasley (2004) provide a detailed discussion of the research
design issues associated with the Mishkin test.

26
SIZEt+1 = 0 + 1I_RMnt + 2SIZEtdec + 3BETAtdec + 4LnBMtdec + 5EPtdec + 6ACCtdec (11)

Where: n = 1,2,3,4

The dependent variables are one-year ahead size adjusted returns. Equation (11) includes

a variety of control variables used in accounting and finance literature as proxies for risk factors

that predict stock returns. Research shows that future abnormal returns are associated with

SIZE, book-to-market (BTM), and systematic risk (BETA). Additionally, the earnings-to-price

ratio (EP) is included to control for the earnings-price anomaly and accruals (ACC) to capture

the accrual anomaly documented by Sloan (1996).19

Following Fama and MacBeth (1973), equation (11) is run annually. The coefficient

estimates reported are the means of the time-series coefficients. To address outliers and so that

the coefficients can be interpreted as returns to a zero investment hedge portfolio, the control

variable are ranked by deciles (0,9) each year and the decile number is divided by nine so each

observation takes the value ranging between zero and one (Rajgopal et al., 2003). The t-statistics

are based on the time-series standard errors of the estimated coefficients.

Results reported in Table 9 indicate that incremental abnormal returns related to R&D

RM (Panel A) and Asset RM (Panel D) persist after controlling for the FamaFrench factors and

the accrual anomaly. There is a negative relation between R&D and Asset RM and future

returns that is statistically significant. R&D RM firm-years are associated with incremental

returns of 7.1%. Similarly, Asset RM firm-years are associated with incremental returns of

8.1%. The negative sign on the coefficients is consistent with the difference in historical and

security-market weightings of the contribution of RM to future earnings documents using the

Mishkin framework.

19
To address potential concerns about industry membership on future returns, particularly high tech industries, I
conduct the following two sensitivity tests. First, I run the regression including controls for industries with more
than 100 firms (based on the Fama-French 48 industries). Second, I include an indicator variable for whether the
firm is classified as a new economy firm, using the classification in Murphy (2003). Unreported results indicate the
results are robust to these additional controls.

27
 Table 9, also, reports a negative association between SG&A RM (Panel B) and

Production RM (Panel D) and future returns. The incremental returns associated with SG&A

RM are negative for nine out of the 13 years. Additionally, incremental returns associated with

Production RM are negative for eight out of the 13 years. However, after controlling for the

FamaFrench factors and the accrual anomaly the incremental returns are not statistically

significant. Therefore, identifying SG&A and Production RM does not appear to be informative

in explaining future returns.

5.3.2 Analyst forecasts

The primary finding is that RM results in an economically significant decline in

operating performance. In this section, I investigate whether the poor earnings performance of

RM firms are a surprise to analysts. Using data provided by I/B/E/S, the forecast error is equal

to actual realized earnings per share minus the mean of the analysts forecasts. To control for

industry-wide surprises and bias in analysts forecasts, I compute the earnings forecast error on a

matched sample using the same criteria as the univariate analysis (performance, industry and

accruals decile). The abnormal analyst forecast error is the RM firms forecast error minus the

contemporaneous forecast error for its matched control firm.

Table 10 presents the results. In general, the results reveal that the analysts forecast

errors for all four types of RM are not statistically different from the control firms. Although

unreported results reveal that forecast errors for the RM sample are significantly negative

implying that analysts are overly optimistic for the RM firms, the abnormal forecast error is not

significantly different from zero for any type of RM.

It appears analysts recognize the future earnings implications of all four types of RM.

This is consistent with the recommendations of financial statement analysis texts that encourage

analysts be attentive to expenditures on R&D, marketing and distribution activities and the firms

overall competitive strategy and to closely examine potential red flags such as realized gains

from asset sales (Palepu, Healy and Bernard, 2004).

28
6. Conclusion

This paper contributes to the body of literature examining the resource allocation effects

of earnings management. Four types of real earnings management activities are examined: (1)

cut discretionary investment of R&D to decrease expense, (2) cut discretionary investment of

SG&A to decrease expense, (3) sell fixed assets to report gains, and (4) cut prices or extend

more lenient credit terms to boost sales and/or overproduce to decrease COGS expense. Next, I

assess: (i) the extent to which real earnings management affects subsequent operating

performance (as measured by both earnings and cash flows), and (ii) whether investors and

analysts expect the subsequent decline in performance.

The analysis illustrates that real earnings management has an economically significant

impact on subsequent operating performance. Specifically, all four types of real earnings

management activities are associated with lower return on assets compared to non-RM firms

after controlling for size, performance, accruals, and industry. Except for myopic investment in

SG&A, the other three types of RM are associated with significantly lower future cash flow

scaled by assets. The regression results indicate that all four types of real earnings management

are associated with lower ROA in the subsequent year controlling for past performance, size,

growth, and accruals decile. In addition, the persistence of ROA is significantly lower for Asset

and Production RM firm-years. The analysis suggests that, overall, identifying all four types of

RM is incrementally informative about future earnings and cash flows.

Given the empirical results that all four types of real earnings management activities

negatively impact future operating performance, I turn my attention the question of whether

investors and analysts recognize the consequences of real earnings management. The analysis

provides evidence that investors expectations, as reflected in stock prices, do not recognize the

consequences of myopic R&D investment and the strategic timing of asset sales but the evidence

is inconsistent with investors not understanding the implications of myopic investment in SG&A

and cutting prices and/or overproducing to increase current period income. Analysts

29
expectation, as reflected in forecasts of earnings, appear to incorporate information about all four

types of real earnings management.

30
References

Altamuro, J., Beatty, A. and Weber, J. (2003) Motives for early revenue recognition: evidence from
SEC staff accounting bulletin (SAB) 101. (working paper, MIT).

Anderson, M., Banker, R.D. and Janakiraman, S.N. (2003) Are selling, general, and administrative
costs sticky? Journal of Accounting Research 41 (1): 47-63.

Baber, W., Fairfield, P.M. and Haggard, J.A. (1991) The effect of concern about reported income on
discretionary spending decisions: the case of research and development. The Accounting Review 66
(4): 818-829.

Barber, B.M., and Lyon, J.D. (1997) Detecting long-run abnormal stock returns: the empirical power
and specification of test statistics. Journal of Financial Economics 43: 341-372.

Barton, J. and Simko, P. (2002) The balance sheet as an earnings management constraint. The
Accounting Review 77: 1-27.

Bartov, E. (1993) The timing of asset sales and earnings manipulation. The Accounting Review. 68
(4): 840-855.

Basu, S. (1977) The investment performance of common stock in relation to their price to earnings
ratios: a test of the efficient markets hypothesis. The Journal of Finance 32: 663-662.

Beaver, W.H. and Engel, E (1996) Discretionary behavior with respect to allowances for loan losses
and the behavior of security price. Journal of Accounting and Economics 22: 177-206.

Beaver, W.H. and McNichols, M.F. (2001) Do stock prices of property casualty insurers fully reflect
information about earnings, accruals, cash flows, and development? Review of Accounting Studies 6:
197-220.

Bens, D., Nagar, V. and Wong, M.H.F. (2002) Real investment implications of employee stock option
exercises. Journal of Accounting Research 40 (2): 359-406.

Berger, P. (1993) Explicit and Implicit tax effects of the R&D tax credit. Journal of Accounting
Research. 31 (2): 131-171.

Bradshaw, M., Richardson, S. and Sloan R. (2001) Do analysts and auditors use information in
accruals? Journal of Accounting Research 39 (1): 45-74.

Bushee, B. (1998) The influence of institutional investors on myopic R&D investment behavior. The
Accounting Review. 73 (3): 305-333.

Cheng, S. (2003) R&D expenditures and CEO compensation. (working paper: University of
Michigan).
Dechow, P., Kothari, S. and Watts, R. (1998) The relation between earnings and cash flows. Journal of
Accounting and Economics. 25 (2): 133-169.

Dechow, P.M. and Skinner, D.J. (2000) Earnings management: Reconciling the views of accounting
academics, practitioners, and regulators. Accounting Horizons 14 (June): 235-250.

Dechow, P. and Sloan, R. (1991) Executive incentives and the horizon problem. Journal of Accounting
and Economics 14: 51-89.

Dechow, P., Sloan, R. and Sweeney, A. (1996) Causes and consequences of earnings manipulation: An
analysis of firms subject to enforcement actions by the SEC. Contemporary Accounting Research 13:
1-36.

DeFond, M. and Park, L. (1997) Smoothing income in anticipation of future earnings. Journal of
Accounting Research, 40 (4): 115-139.

DeFond, M. (2002) Discussion of the balance sheet as an earnings management constraint. The
Accounting Review 77: 29-33.

Ewert, R. and Wagenhofer, A. (2004) Economic effects of tightening accounting standards to restrict
earnings management. (working paper: Goethe-University, Frankfurt Germany and University of Graz,
Austria).

Fama, E. and French, K. (1997) Industry costs of equity. Journal of Financial Economics 43: 153-193.

Fama, E. and Macbeth, J. (1973) Risk, return and equilibrium: empirical tests. Journal of Political
Economy 71, 607-636.

Fischer, P. and Verrecchia, R. (2000) Reporting Bias. The Accounting Review 75 (2): 229-245.

Francis, J., LaFond, R., Olsson, P. and Schipper, K. (2003) Earnings quality and the pricing effects of
earnings patterns. (working paper: Duke, University of Wisconsin and FASB).

Graham, J., Harvey, R. and Rajgopal, S. (2004) The economic implications of corporate financial
reporting. (working paper: Duke, NBER and University of Washington, Seattle).

Hansen, J. (2004). Why small loss firms? Earnings management constraints and market senditivity to
earnings. (working paper: University of Illinois at Chicago).

Herrmann, T., Inoue, T. and Thomas, W.B. (2003) The sale of assets to manage earnings in Japan.
Journal of Accounting Research 41 (1) 89-108.

Holthausen, R., Larcher, D.F. and Sloan, R.G. (1995) Annual bonus schemes and the manipulation of
earnings. Journal of Accounting and Economics 19: 29-74.

Hribar, P. and Collins, D. (2002) Errors in estimating accruals: implications for empirical research.
Journal of Accounting Research 40 (1): 105-134.

2
Jones, J. (1991) Earnings management during import relief investigations. Journal of Accounting 29
(2): 193-228.

Kasznik, R. (1999) On the association between voluntary disclosure and earnings management. Journal
of Accounting Research 37 (1): 57-81.

Kraft, A., Leone, A. and Wasley, C. (2004) On the use of mishkins rational expectations approach to
testing efficient-markets hypotheses in accounting research. (working paper, University of Rochester).

Kothari, S., Leone, A., and Wasley, C. (2002) Performance matched discretionary accrual measures.
(working paper, MIT).

Kothari, S. and Sloan, R. (1992) Information in Prices About Future Earnings: Implications for
Earnings Response Coefficients. Journal of Accounting and Economics. 15 (2) 143-172.

Lev, B. and Sougiannis, T. (1996) The capitalization, amortization and value-relevance of R&D.
Journal of Accounting and Economics (1996) 21 (1): 107-139.

Liu, C., Ryan, S. and Wahlen, J. (1997) Differential valuation implications of loan loss provisions
across banks and fiscal quarters. The Accounting Review. 72 (1): 133-146.

Mishkin, F. (1983) A rational expectations approach to macroeconometrics: testing policy

effectiveness and effects markets models. Chicago, Illinois: University of Chicago Press for the
National Bureau of Economic Research.

Murphy, K. (2003) Stock based pay in new economy firms. Journal of Accounting and Economics. 34
(3): 129-147.

Palepu, K.G., Healy, P.M., and Bernard, V.L. (2004) Business Analysis and Valuation. 3rd Edition.
South Western College Publishing.

Penman S. and Zhang X. (2002) Accounting conservatism, the quality of earnings, and stock returns.
The Accounting Review 77 (2): 237-263.

Perry, S. and Williams, T. (1994) Earnings management preceding management buyout offers. The
Accounting Review 18 (2): 157-180.

Rajgopal, S., Shevlin, T. and Venkatachalam, M. (2003) Does the stock market fully appreciate the
implications of leading indicators for future earnings? Evidence from order backlog. Review of
Accounting Studies 8: 461-492.

Roychowdhury, S. (2003) Management of earnings through the manipulation of real activities that
affect cash flow from operations. (working paper, MIT).
Schipper, K. (1989) Commentary on earnings management. Accounting Horizons. (3): 91-102.

Sloan, R. (1996) Do stock prices fully reflect information in accruals and cash flows about future
earnings? The Accounting Review 71 (July): 289-281.

Subramanyam, K.R. (1996) The price of discretionary accruals. Journal of Accounting and Economics
22: 249-281.

Teoh, S., Wong T. and Rao G. (1998) Are earnings during initial public offerings opportunistic?
Review of Accounting Studies 3.

Teoh, S., Welch, I. and Wong, H. (1998a) Earnings management and the long-run underperformance
of initial public offerings. Journal of Finance 53 (6): 1935-1974.

Thomas, J. K. and Zhang H. (2002) Inventory changes and future returns. Review of Accounting
Studies 7: 163-187.

Wahlen, J. (1994) The nature of information in commercial bank loan loss disclosures. The
Accounting Review 69 (July): 455-478.

Watts, R. and Zimmerman, J. (1978) Towards a positive theory of the determination of accounting
standards. The Accounting Review 53 (January): 112-134.

Xie, H. (2001) The mispricing of abnormal accruals. The Accounting Review 76 (July): 356-373.

Zang, A. (2003) When do managers use real activities manipulation? (working paper, Duke
University).
 Table 1
Mean and Median Prior Cumulative Abnormal Accruals across Quintiles of Net Operating Assets.
Sample consists of 32,402 firm-years from 1988-2000. Replication of Barton and Simko (2002), shows that firms with
higher net operating assets (NOA) are associated with higher levels of previous accruals management. NOA is net operating
assets (i.e. shareholders' equity less cash and marketable securities, plus total debt) at the beginning of year t, scaled by sales
for year t-1. Abnormal accruals are estimated for firm i in year t using the residual from the Jones (1991) model, estimated
by two-digit SIC code. Abnormal accruals are implicitly scaled by lagged total assets, therefore before accumulating
abnormal accruals back in time, they are unscaled by multiplying the regression residuals by the corresponding lagged total
assets. To avoid inducing a spurious correlation between NOA and prior cumulative abnormal accruals, abnormal accruals
are scaled by lagged sales, the same deflator in NOA.

Difference between
Quintiles of NOA (at the beginning of the year) by year and industry
Quintile 5 and 1
1 (lowest) 2 3 4 5 (highest) (p-value)

Panel A: Mean Abnormal Accruals

Accumulation Period (years):

[t-1 , t] -0.09 -0.04 0.00 -0.04 -0.01 0.07
[t-2 , t] -0.15 -0.09 -0.02 -0.05 -0.04 0.01
[t-3 , t] -0.22 -0.12 -0.03 -0.09 -0.06 0.16
[t-4 , t] -0.26 -0.17 -0.06 -0.12 -0.08 0.18
[t-5 , t] -0.32 -0.21 -0.07 -0.13 -0.13 0.20

Panel B: Median Abnormal Accruals

Accumulation Period (years):

[t-1 , t] -0.010 -0.003 0.001 0.007 0.014 0.04
[t-2 , t] -0.020 -0.007 0.002 0.010 0.024 0.06
[t-3 , t] -0.029 -0.012 0.003 0.013 0.030 0.08
[t-4 , t] -0.037 -0.013 0.002 0.016 0.033 0.09
[t-5 , t] -0.047 -0.015 0.005 0.017 0.035 0.09

The variables are defined as follows:

NOA = net operating assets (shareholders' equity less cash and marketable securities), plus total debt divided by lagged assets ((#216- #1 + #9 +34)
Mean/Median Abnormal Accruals estimated using the following the modified Jones model:

ACC t 1 REV t REC t PPE t

= 0 + 1 + 2 +
At 1 At 1 At 1 At 1
ACC = income before extraordinary items less net cash flows from operating activities divided by lagged total assets (#123-#308)
A = total assets #6
REV = change in sales #12
REC = change in total receivables #2
PPE = property plant and equipment #7
 Table 2
Identification of RM Firm-Years

Frequency of firms by year t abnormal RM quintile measure and year t-1 net operating asset (NOA) quintile. Total sample
32,484 firm-years from 1988 to 2000. NOA is net operating assets (i.e. shareholders' equity less cash and marketable securities,
plus total debt) at the beginning of year t, scaled by sales for year t-1. R&D RM firm-years are those in the lowest abnormal
R&D expense quintile and the highest NOA quintile. SG&A RM firm-years are those in the lowest abnormal SG&A expense
quintile and the highest NOA quintile. Asset RM firm-years are those in the highest abnormal gain on sale of fixed asset quintile
and the highest NOA quintile. Production RM firm-years are those in the highest abnormal production cost quintile and highest
NOA quintile.

Panel A: Frequency by Quintile of Abnormal R&Dt and Quintile of NOAt-1 Descriptive Statistics

Number of Firms by NOA Quintile in year t-1

Abnormal R&D Mean Abnormal
Quintile R&D 1 2 3 4 5 Accruals NOAt-1

1 (highest) 0.15 980 837 739 649 661 -0.078 0.984

2 0.01 798 844 862 774 598 -0.042 0.681
3 0.00 691 871 803 837 675 -0.040 0.715
4 -0.03 643 782 872 851 728 -0.034 0.797
5 (lowest) -0.13 674 710 743 768 976 -0.038 1.422

Panel B: Frequency by Quintile of Abnormal SG&At and Quintile of NOAt-1

Number of Firms by NOA Quintile in year t-1

Abnormal SG&A Mean Abnormal
Quintile SG&A 1 2 3 4 5 Accruals NOAt-1

1 (highest) 0.25 939 884 951 1,056 1,120 -0.054 1.083

2 0.05 932 1,056 1,061 1,134 777 -0.039 0.748
3 0.00 977 1,115 1,129 1,018 717 -0.035 0.683
4 -0.06 981 1,171 1,101 944 763 -0.039 0.729
5 (lowest) -0.24 1,059 996 912 904 1,080 -0.047 1.149

Panel C: Frequency by Quintile of Abnormal Gain on Asset Salest and Quintile of NOAt-1

Abnormal Gain on Mean Abnormal

Number of Firms by NOA Quintile in year t-1
Asset Sale Gain on Asset
Quintile Sales 1 2 3 4 5 Accruals NOAt-1

1 (lowest) -0.41 854 905 895 968 793 -0.061 0.960

2 -0.08 826 969 957 922 752 -0.050 0.875
3 -0.03 871 964 968 860 765 -0.044 0.909
4 0.01 869 864 880 897 916 -0.036 1.029
5 (highest) 0.45 805 786 813 866 1,150 -0.040 1.258
 Table 2 (cont.)
Identification of RM Firm-Years

Panel D: Frequency by Quintile of Abnormal Production Costst and Quintile of NOAt-1

Descriptive Statistics
Abnormal
Number of Firms by NOA Quintile in year t-1
Production Cost Mean Abnormal
Quintile Production Costs 1 2 3 4 5 Accruals NOAt-1

1 (lowest) -0.45 1,225 1,339 1,314 1,251 890 -0.032 0.661

2 -0.15 874 1,143 1,329 1,372 1,308 -0.042 0.996
3 -0.04 806 1,087 1,251 1,387 1,500 -0.048 1.211
4 0.06 1,054 1,279 1,185 1,268 1,240 -0.050 0.998
5 (highest) 0.38 1,961 1,303 1,041 820 898 -0.058 1.083

The variables are defined as follows:

Accruals = income before extraordinary items less net cash flows from operating activities divided by average total assets
NOA = beginning balance of net operating assets (shareholders' equity less cash and marketable securities, plus total debt)
Estimation Models to Calculate Abnormal Level of R&D, SG&A, Gain on Asset Sale and Production Costs:
1. Abnormal R&D is the residual from the following model estimated by year and industry:

RDt RDt 1
= 0 + 1 + 2 INTt + 3Qt + 4CXt + 5 MVt +
At 1 At 2

2. Abnormal SG&A is the residual from the following model estimated by year:
SGAt S S S S
log( ) = 0 + 1 log( t ) + 2 log( t ) * DDt + 3 log( t 1 ) + 4 log( t 1 ) * DDt 1 +
SGAt 1 S t 1 S t 1 S t 2 S t 2
3. Abnormal Gain on Asset Sales is the difference between the actual gain minus the industry year median

GainAt = 0 + 1 ASalest + 2 ISalest + 3 log(S ) t + 4 % S t

4. Abnormal Production Costs is the residual from the following model estimated by year and industry:
PRODt 1 S S t S
= 0 + 1 t + 2 + 3 t 1 +
At 1 At 1 At 1 At 1 At 1
RD = research and development expense
A = total assets
INT = internal funds (income before extraordinary items plus research and development expense plus depreciation and amortortization) divided by total assets
Q = tobin's Q (market value of equity plus preferred stock plus long term debt plus debt in current liabilities) divided by assets
CX = capital expenditure divided by assets
MV = log of the market value of equity
SGA = selling, general and administrative expense and advertising plus advertising expense
S = total sales
DD = indicator variable equal to 1 when sales decreases between t-1 and t, zero otherwise
ASales =long-lived assets sales divided by the market value at the beginning of the year
ISales = Long-lived investment sales / market value at the beginning of the year [data109/(#199*#25)]
PROD = COGS + INV
COGS = cost of goods sold expense
INV = inventory increase/decrease
 Table 3
Descriptive Statistics

Firm-years from 1988-2000. The full sample consists of all firm-years (excluding the RM firm-years). The real
earnings management sample consists of all firm-years identified as having engaged in one or more types of RM.
Of the 3,129 RM firm-years: 2,280, 730, 112 and 7 engage in one, two, three and four types of RM, respectively.
NOA is net operating assets (i.e. shareholders' equity less cash and marketable securities, plus total debt) at the
beginning of year t, scaled by sales for year t-1. R&D RM firm-years are those in the lowest abnormal R&D
expense quintile and the highest NOA quintile. SG&A RM firm-years are those in the lowest abnormal SG&A
expense quintile and the highest NOA quintile. Asset RM firm-years are those in the highest abnormal gain on
sale of fixed asset quintile and the highest NOA quintile. Production RM firm-years are those in the highest
abnormal production cost quintile and highest NOA quintile.

Variable Mean Std.dev. Median 1st quartile 3rd quartile

Full Sample (excluding RM firm-years, n=29,355)

TA 1,331 8,282 135 42 561
TS 1,407 6,278 160 43 668
MV 1,808 11,335 143 39 664
ROA 0.02 0.17 0.05 0.00 0.10
CFO 0.06 0.16 0.06 0.02 0.14
ACC -0.05 0.12 -0.04 -0.09 0.01
SARt+1 0.05 0.63 -0.06 -0.34 0.26

RM Sample (n=3,129)
TA 1,243 7,778 74 24 285
TS 772 4,949 44 10 186
MV 902 4,346 98 32 353
ROA -0.10 0.27 0.01 -0.18 0.06
CFO -0.04 0.22 0.02 -0.10 0.09
ACC -0.06 0.16 -0.04 -0.11 0.01
SARt+1 0.00 0.71 0.00 -0.45 0.22
The variables are defined as follows:
TA = total assets in millions
TS = total sales in millions
MV = market value of equity in millions
ROA = income before extraordinary items divided by total assets
CFO = cash flows from operations divided by total assets
ACC = income before extraordinary items minus cash flows from operations divided by average total assets
SAR = size adjusted abnormal returns computed as the buy and hold raw retun minus the buy and hold return on a size matched decile portfolio of
firms cumulated over 12 months beginning with the fourth month after the end of fiscal year t.
 Table 3 (cont.)
Descriptive Statistics
RM Samples categorized by industry. Firms are assigned to 48 industries based on the classification system
developed by Fama and French (1997). NOA is net operating assets (i.e. shareholders' equity less cash and
marketable securities, plus total debt) at the beginning of year t, scaled by sales for year t-1. R&D RM firm-years
are those in the lowest abnormal R&D expense quintile and the highest NOA quintile. SG&A RM firm-years are
those in the lowest abnormal SG&A expense quintile and the highest NOA quintile. Asset RM firm-years are
those in the highest abnormal gain on sale of fixed asset quintile and the highest NOA quintile. Production RM
firm-years are those in the highest abnormal production cost quintile and highest NOA quintile.

R&D RM SG&A RM Assets RM Production

Industry Full Sample
firms firms firms RM firms
Business Services 4,266 224 154 136 133
Electronic Equipment 2,663 136 106 88 46
Retail 2,189 7 61 106 37
Pharmaceutical Products 2,016 202 27 40 230
Computers 1,925 141 82 73 56
Petroleum and Natural Gas 1,685 93 82 30
Medical Equipment 1,542 133 65 27 56
Wholesale 1,525 4 60 44 24
Machinery 1,488 28 41 60 39
Measuring and Control Equip. 1,125 76 46 41 39
Construction Materials 912 1 35 41 9
Chemicals 868 5 24 38 20
Consumer Goods 764 1 32 21 27
Electrical Equipment 693 13 25 42 16
Other 8,823 5 229 311 136
Total 32,484 976 1,080 1,150 898
 Table 4
Abnormal Operating Performance for RM Firms in the Subsequent Three Years
Firm-years suspected of RM are matched to control firms by industry membership, accruals decile and performance
(within 10%). NOA is net operating assets (i.e. shareholders' equity less cash and marketable securities, plus total
debt) at the beginning of year t, scaled by sales for year t-1. Panel A contains firm-years for which firms are suspected
of R&D RM (lowest abnormal R&D expense quintile and highest NOA quintile). Panel B contains firm-years for
which firms are suspected of SG&A RM (lowest abnormal SG&A expense quintile and highest NOA quintile). Panel
C contains firm-years for which firms are suspected of Asset RM (highest gain on sale of fixed asset quintile and
highest NOA quintile). Panel D contains firm-years for which firms are suspected of Production RM (highest
production cost quintile and highest NOA quintile).
Abnormal ROAi,t = ROAi,t Matched ROAi,t (5)
Abnormal CFOi,t = CFOi,t Matched CFOi,t (6)
Abnormal Operating Performance (%) p-value for Difference
N Mean Median Mean Median

Panel A: R&D RM
ROA t 851 -0.06 0.00 0.15 0.38
t+1 851 -4.19*** -1.15*** 0.00 0.00
t+2 791 -3.64*** -1.53*** 0.00 0.00
t+3 734 -1.44 -0.65* 0.19 0.08

CFO t 834 -0.02 0.00 0.57 0.66

t+1 834 -1.43** -1.47*** 0.03 0.01
t+2 778 -1.58** -1.12 0.05 0.11
t+3 725 -1.08 -1.23* 0.21 0.10

Panel B: SG&A RM
ROA t 964 -0.01 0.00 0.79 0.46
t+1 964 -1.96*** -0.31** 0.01 0.03
t+2 909 -2.03*** -1.02*** 0.01 0.00
t+3 846 -2.59*** -0.22** 0.01 0.05

CFO t 951 -0.01 0.01 0.75 0.42

t+1 949 -0.09 -0.44 0.87 0.32
t+2 888 -0.59 -1.01 0.35 0.20
t+3 830 -0.53 -0.22 0.47 0.50

Panel C: Asset RM
ROA t 1021 -0.03 0.00 0.37 0.83
t+1 1021 -0.57 -0.20 0.44 0.20
t+2 966 -1.23 -0.32* 0.11 0.06
t+3 910 -1.70* -0.36** 0.06 0.05

CFO t 1006 -0.01 0.00 0.69 0.99

t+1 1005 -0.73 -0.63*** 0.18 0.02
t+2 954 -0.75 -1.09* 0.21 0.10
t+3 899 -0.15 -0.05 0.84 0.53

Panel D: Production RM
ROA t 763 -0.03 0.00 0.61 0.33
t+1 763 -3.26*** -1.27*** 0.00 0.00
t+2 723 -3.76*** -2.49*** 0.00 0.00
t+3 684 -2.67** -1.91*** 0.03 0.00

CFO t 722 -0.10* -0.01* 0.06 0.10

t+1 722 -1.79** -1.46*** 0.03 0.01
t+2 683 -1.44 -0.92 0.14 0.16
t+3 647 -0.03 -0.80 0.98 0.41
 Table 5
Abnormal Change in Operating Performance for RM Firms in the Subsequent Three Years
Firm-years suspected of RM are matched to control firms by industry membership, accruals decile and performance
(within 10%). NOA is net operating assets (i.e. shareholders' equity less cash and marketable securities, plus total debt) at
the beginning of year t, scaled by sales for year t-1. Panel A contains firm-years for which firms are suspected of R&D
RM (lowest abnormal R&D expense quintile and highest NOA quintile). Panel B contains firm-years for which firms are
suspected of SG&A RM (lowested abnormal SG&A expense quintile and highest NOA quintile). Panel C contains firm-
years for which firms are suspected of Asset RM (highest gain on sale of fixed asset quintile and highest NOA quintile).
Panel D contains firm-years for which firms are suspected of Production RM (highest production cost quintile and highest
NOA quintile).
Abnormal ROAi,t = [ROAi,t ROAi,t-1] [Matched ROAi,t Matched ROAi,t-1]
Abnormal CFOi,t = [CFOi,t CFOi,t-1] [Matched CFOi,t Matched CFOi,t-1]

Abnormal Operating Performance (%) p-value for Difference

N Mean Median Mean Median

Panel A: R&D RM
ROA t+1 851 -4.12*** -1.37*** 0.00 0.00
t+2 791 0.37 -0.83 0.70 0.46
t+3 734 1.51 0.15 0.15 0.50

CFO t+1 834 -1.41** -1.51*** 0.04 0.01

t+2 778 -0.41 0.57 0.57 0.90
t+3 725 0.29 -0.25 0.69 0.82

Panel B: SG&A RM
ROA t+1 964 -1.95*** -0.35** 0.01 0.03
t+2 909 -0.12 -0.36 0.88 0.36
t+3 843 -0.26 0.11 0.77 0.72

CFO t+1 949 -0.08 -0.25 0.88 0.33

t+2 887 -0.51 -0.25 0.41 0.93
t+3 826 0.09 0.06 0.89 0.98

Panel C: Asset RM
ROA t+1 1021 -0.54 -0.25 0.46 0.19
t+2 966 -0.60 -0.11 0.45 0.55
t+3 909 -0.14 -0.16 0.87 0.53

CFO t+1 1005 -0.72 -0.65** 0.18 0.02

t+2 953 -0.08 -0.25 0.89 0.66
t+3 899 0.48 0.29 0.45 0.34

Panel D: Production RM
ROA t+1 763 -3.23*** -1.33*** 0.00 0.00
t+2 723 -0.30 -0.76 0.79 0.38
t+3 682 1.25 -0.11 0.26 0.73

CFO t+1 722 -1.69** -1.33*** 0.04 0.01

t+2 683 0.24 0.06 0.78 0.83
t+3 645 0.99 0.05 0.29 0.42
 Table 6
Cross-Sectional Regressions Relating ROAt+1 to RM and Control Variables

Coefficient estimates of ordinary least squares regression relating ROAt+1 to an indicator variable for RM firms, an interactive term equal to
ROA multiplied by the RM indicator variable and control variables. Past ROA controls for past performance and LOGASSET controls for any
size effects. BTM controls for growth opportunities and/or the life cycle of the firm. RETURN controls for the association between stock
performance and future earnings. PORTACC is included to control for the accrual anomaly documented by Sloan (1996). Sample consists of
firm-years from 1988-2000 in the highest net operating asset quintile.

ROAt+1= 0 + 1LOGASSETt + 2BTMt + 3ROAt + 4RETURNt + 5PORTACCt + 6I_RMnt + 7I_RMnt*ROAt + t+1 (7)
where n = 1,2,3,4

Model 1: R&D Model 2: SG&A Model 3: Asset Model 4: Production

Variable Pred. Sign INDRM1 INDRM2 INDRM3 INDRM4
Intercept ? -0.077 -0.046 -0.051 -0.055
(-6.36)*** (-5.12)*** (-5.47)*** (-6.77)***

LOGASSET 0.012 0.006 0.009 0.010

(7.68)*** (5.24)*** (7.31)*** (8.71)***

BTM 0.006 -0.004 -0.004 -0.004

(1.24) (-5.25)*** (-4.89)*** (-4.94)***

ROA + 0.561 0.548 0.639 0.579

(46.76)*** (47.92)*** (55.11)*** (57.38)***

RETURN + 0.017 0.023 0.019 0.018

(6.82)*** (10.85)*** (8.23)*** (8.81)***

PORTACC -0.034 -0.032 -0.054 -0.037

(-3.42)*** (-4.16)*** (-6.68)*** (-5.19)***

I_RMn -0.044 -0.009 -0.005 -0.058

(-6.33)*** (-1.62)* (-0.92) (-8.92)***

I_RMn * ROA 0.025 0.015 -0.027 -0.047

(2.13)** (1.20) (-2.51)** (-5.59)***

N 3,399 4,112 4,075 5,384

2
Adj. R 55.1% 49.4% 53.8% 53.1%
*/**/*** represent statistical significance at 10%/5%/1% levels two-tailed. t-values in parentheses.
The variables are defined as follows:
ROA = income before extraordinary items divided by total assets
LOGASSET = the natural logarithm of total assets
BTM = the book value of equity divided by the market value of equity
RETURN = the one year holding period return on an investment in the common stock of the firm
PORTACC = the portfolio ranking of accruals divided by total assets, converted to a [0,1] scale, with 0 (1) representing the lowest (highest) level of accruals
I_RM1 = R&D RM: an indicator variable equal to one if the firm is in the lowest abnormal R&D expense quintile and belongs to the highest quintile of beginning balance of NOA
I_RM2 = SG&A RM: an indicator variable equal to one if the firm is in the lowest abnormal SG&A expense quintile and belongs to the highest quintile of beginning balance of NOA
I_RM3 = Asset RM: an indicator variable equal to one if the firm is in the highest abnormal gain on assets sale quintile and belongs to the highest quintile of beginning balance of NOA
I_RM4 = Production RM: an indicator variable equal to one if the firm is in the lowest abnormal production cost quintile and belongs to the highest quintile of beginning balance of NOA
 Table 7
Cross-Sectional Regressions Relating ROAt+2 to RM and Control Variables
Coefficient estimates of ordinary least squares regression relating ROAt+1 to an indicator variable for RM firms, an interactive term equal to
ROA multiplied by the RM indicator variable and control variables. Past ROA controls for past performance and LOGASSET controls for any
size effects. BTM controls for growth opportunities and/or the life cycle of the firm. RETURN controls for the association between stock
performance and future earnings. PORTACC is included to control for the accrual anomaly documented by Sloan (1996). Sample consists of
firm-years from 1988-2000 in the highest net operating asset quintile.

ROAt+2= 0 + 1LOGASSETt + 2BTMt + 3ROAt + 4RETURNt + 5PORTACCt + 6I_RMnt + 7I_RMnt*ROAt + t+1

where n = 1,2,3,4

Model 1: R&D Model 2: SG&A Model 3: Asset Model 4: Production

Variable Pred. Sign I_RM1 I_RM2 I_RM3 I_RM4
Intercept ? -0.088 -0.056 -0.058 -0.057
(-6.64)*** (-5.67)*** (-5.50)*** (-6.40)***

LOGASSET /+ 0.012 0.008 0.011 0.010

(7.03)*** (6.12)*** (7.57)*** (8.28)***

BTM 0.022 0.000 0.000 0.000

(4.04)*** (-0.36) (-0.47) (-0.11)

ROA + 0.467 0.397 0.493 0.431

(35.20)*** (30.96)*** (36.20)*** (38.16)***

RETURN + 0.017 0.020 0.014 0.017

(6.17)*** (8.32)*** (5.39)*** (7.59)***

PORTACC -0.035 -0.023 -0.052 -0.038

(-3.22)*** (-2.75)*** (-5.60)*** (-4.88)***

I_RMn -0.033 -0.012 -0.001 -0.051

(-4.33)*** (-2.06)** (-0.08) (-7.05)***

I_RMn * ROA -0.016 0.045 -0.011 0.011

(-1.22) (3.16)*** (-0.74) (1.23)

N 3,399 4,112 4,075 5,384

2
Adj. R 0.42 0.33 0.37 0.39
***/**/* represent statistical significance at 10%/5%/1% levels two-tailed. t-values in parentheses.
The variables are defined as follows:
ROA = income before extraordinary items divided by total assets
LOGASSET = the natural logarithm of total assets
BTM = the book value of equity divided by the market value of equity
RETURN = the one year holding period return on an investment in the common stock of the firm
PORTACC = the portfolio ranking of accruals divided by total assets, converted to a [0,1] scale, with 0 (1) representing the lowest (highest) level of accruals
I_RM1 = R&D RM: an indicator variable equal to one if the firm is in the lowest abnormal R&D expense quintile and belongs to the highest quintile of beginning balance of NOA
I_RM2 = SG&A RM: an indicator variable equal to one if the firm is in the lowest abnormal SG&A expense quintile and belongs to the highest quintile of beginning balance of NOA
I_RM3 = Asset RM: an indicator variable equal to one if the firm is in the highest abnormal gain on assets sale quintile and belongs to the highest quintile of beginning balance of NOA
I_RM4 = Production RM: an indicator variable equal to one if the firm is in the lowest abnormal production cost quintile and belongs to the highest quintile of beginning balance of NOA
 Table 8
The Mishkin Test
Firm-years from 1988-2000. Nonlinear generalized least squares estimates of the market pricing of cash flows and accruals with respect to their
implications for one-year ahead earnings. The likelihood ratio statistic tests the equality of coefficients across equation (10) and equation (11).

"Forecasting Equation"
EARNt+1 = 0 + 1aCFOt + 1bACCt + 2I_RMnt + 2aCFOt*I_RMnt + 2bACCt*I_RMnt +t+1 (8)
"Pricing Equation"
SARt+1 = 0 + 1 (EARNt+1 0 1a*CFOt 1b*ACCt + 2*I_RMnt 2a*CFOt*I_RMnt 2b*ACCt*I_RMnt) + t+1 (9)

Equation (10) Equation (11) Likelihood Marginal

ratio significance
Variable Pred. Sign Coefficient t-statistic Coefficient t-statistic statistic level

Panel A: R&D RM: I_RM1 (n = 19,360)

Intercept ? 0.002 2.28*** 0.058 11.44***
CFO + 0.843 179.32*** 0.766 36.13***
ACC + 0.686 88.43*** 0.871 24.68***
I_RM2 -0.026 -6.55*** 0.024 1.31 7.29 0.01
CFO * I_RM2 0.010 0.61 0.118 1.65* 2.12 0.15
ACC * I_RM2 0.097 3.26*** 0.115 0.86 0.00 1.00

Panel B: SG&A RM: I_RM2 (n = 24,769)

Intercept ? 0.009 11.55*** 0.031 6.75***
CFO + 0.806 170.36*** 0.685 36.07***
ACC + 0.660 108.78*** 0.754 31.29***
I_RM2 -0.008 -2.36** -0.005 -0.35 0.26 0.61
CFO * I_RM2 -0.008 -0.50 0.152 2.39** 5.97 0.01
ACC * I_RM2 -0.055 -2.31** -0.056 -0.59 0.00 1.00

Panel C: Asset RM: I_RM3 (n = 22,105)

Intercept ? 0.004 5.12*** 0.050 10.57***
CFO + 0.841 183.31*** 0.786 37.97***
ACC + 0.686 102.68*** 0.853 28.07***
I_RM3 -0.008 -2.39** 0.034 2.28** 7.46 0.01
CFO * I_RM3 0.014 0.82 -0.066 -0.88 1.00 0.32
ACC * I_RM3 0.062 2.21** -0.049 -0.39 0.75 0.39

Panel D: Production RM: I_RM4 (n = 30,114)

Intercept ? 0.006 9.08*** 0.042 10.45***
CFO + 0.822 208.08*** 0.760 42.93***
ACC + 0.688 120.98*** 0.841 32.71***
I_RM4 -0.021 -4.80*** 0.032 1.67* 7.23 0.01
CFO * I_RM4 0.075 5.26*** 0.170 2.66*** 2.07 0.15
ACC * I_RM4 -0.169 -7.18*** -0.376 -3.58*** 3.62 0.06

*/**/*** represent statistical significance at 10%/5%/1% levels two-tailed.

The variables are defined as follows:
EARN = net income before extraordinary items divided by average total assets
CFO = cash flows from operations divided by total assets
ACC = income before extraordinary items minus cash flows from operations divided by average total assets
I_RM1 = R&D RM: an indicator variable equal to one if the firm is in the lowest abnormal R&D expense quintile and belongs to the highest quintile of beginning balance of NOA
I_RM2 = SG&A RM: an indicator variable equal to one if the firm is in the lowest abnormal SG&A expense quintile and belongs to the highest quintile of beginning balance of NOA
I_RM3 = Asset RM: an indicator variable equal to one if the firm is in the highest abnormal gain on assets sale quintile and belongs to the highest quintile of beginning balance of NOA
I_RM4 = Production RM: an indicator variable equal to one if the firm is in the lowest abnormal production cost quintile and belongs to the highest quintile of beginning balance of NOA
SAR = size adjusted abnormal returns computed as the buy and hold raw retun minus the buy and hold return on a size matched decile portfolio of firms cumulated over 12 months beginning with the fourth
month after the end of fiscal year t.
 Table 9
Regressions Relating one year ahead Size Adjusted Returns to to RM and Other Risk Factors

Firm-years from 1988-2000. Summary regression statistics of the relation between abnormal size adjusted stock returns and an RM indicator
variable after controlling for Fama-French risk factors, EP anomaly and Accruals decile (Fama and Macbeth, 1973 approach).

SIZEt+1 = 0 + 1I_RMnt + 2SIZEtdec + 3BETAtdec + 4LnBMtdec + 5EPtdec + 6ACCtdec (10)

Where n = 1,2,3,4

Intercept INDREM SIZEdec BETAdec LnBMdec EPdec ACCdec

? + + +
Panel A: R&D RM: I_RM1
Means from Annual Regressions (N=13) 0.159 -0.071 -0.120 0.046 0.056 -0.079 -0.116
(2.13)* (-3.03)*** (-1.88)* (0.49) (0.65) (-1.17) (-6.61)***
Number of years coefficient Positive/Negative 11/2 3/10 2/11 6/7 7/6 6/7 0/13

Panel B: SG&A RM: I_RM2

Means from Annual Regressions (N=13) 0.103 -0.044 -0.107 0.028 0.064 -0.027 -0.100
(2.64)** (-1.66) (-1.95)* (0.36) (0.89) (-0.55) (-4.01)***
Number of years coefficient Positive/Negative 10/3 4/9 2/11 7/6 7/6 6/7 1/12

Panel C: Asset RM: I_RM3

Means from Annual Regressions (N=13) 0.136 -0.081 -0.134 0.018 0.057 -0.052 -0.097
(2.49)** (-2.42)** (-2.34)** (0.20) (0.69) (-0.85) (-5.64)***
Number of years coefficient Positive/Negative 11/2 4/9 2/11 7/6 9/4 7/6 0/13

Panel D: Production RM: I_RM4

Means from Annual Regressions (N=13) 0.124 -0.057 -0.133 0.041 0.053 -0.040 -0.114
(2.59)** (-1.14) (-2.22)** (0.48) (0.71) (-0.69) (-5.32)***
Number of years coefficient Positive/Negative 10/3 5/8 2/11 7/6 9/4 7/6 1/12

*/**/*** represent statistical significance at 10%/5%/1% levels two-tailed.

SIZEdec = the natural logarithm of market value of common equity measured at the beginning of the abnormal return accumulation period, transformed to a scaled-decile variable with values
ranging from 0 to 1
BETA dec = systematic risk estimated from regression of monthly raw returns on value weighted portfolio over a 60-month return period prior to the abnormal return accumulation period,
transformed to a scaled-decile variable with values ranging from 0 to 1
LnBMdec = the naturnal logarithm of the ratio of the book to market ratio measured at the beginning of the abnormal return accumulation period, transformed ot a scaled-decile variable with values
ranging from 0 to 1
EPdec = earnings to price ratio (stock price measured at the beg. of the return accumulation period), transformed to a scaled-decile variable with values ranging from 0 to 1
dec
ACC = the portfolio ranking of accruals, converted to a [0,1] scale, with 0 (1) representing the lowest (highest) level of accruals
I_RM1 = R&D RM: an indicator variable equal to one if the firm is in the lowest abnormal R&D expense quintile and belongs to the highest quintile of beginning balance of NOA
I_RM2 = SG&A RM: an indicator variable equal to one if the firm is in the lowest abnormal SG&A expense quintile and belongs to the highest quintile of beginning balance of NOA
I_RM3 = Asset RM: an indicator variable equal to one if the firm is in the highest abnormal gain on assets sale quintile and belongs to the highest quintile of beginning balance of NOA
I_RM4 = Production RM: an indicator variable equal to one if the firm is in the lowest abnormal production cost quintile and belongs to the highest quintile of beginning balance of NOA
SAR = size adjusted abnormal returns computed as the buy and hold raw retun minus the buy and hold return on a size matched decile portfolio of firms cumulated over 12 months beginning
with the fourth month after the end of fiscal year t.
 Table 10
Abnormal Analyst Earnings Forecast Errors

Mean and median analysts' forecast errors in the three years after the RM year. For each RM firm, the
forecast error is equal to actual earnings per share minus the median of analysts' forecasts. Firm-years
suspected of RM are matched to control firms by performance, industry membership and accruals decile.
The forecast error for each matched control firm is calculated similarly. The abnormal earnings forecast
error is the difference between the forecast errors for the RM firm and its matched control firm.

Abnormal Forecast Error (%) p-value for Difference

N Mean Median Mean Median

Panel A: R&D RM
t+1 540 -3.26 0.00 0.14 0.23
t+2 465 4.18* 0.00 0.08 0.74
t+3 360 1.99 0.00 0.45 0.83
Panel B: SG&A RM
t+1 466 -2.89 0.00 0.22 0.31
t+2 406 -0.34 0.00 0.88 0.60
t+3 307 1.35 -1.00 0.72 0.31
Panel C: Asset RM
t+1 493 -0.59 0.00 0.84 0.96
t+2 424 -1.96 -1.00* 0.65 0.06
t+3 346 3.13 -1.50 0.46 0.30
Panel D: Production RM
t+1 375 -3.72 0.00 0.24 0.28
t+2 322 -10.24** -1.00 0.04 0.36
t+3 252 -0.84 -1.00 0.80 0.53